I almost never start/stop this machine manually: the proxy wakes it up on incoming TCP connections, and if it’s idle for 600s, the main process exits itself.
Tried again 10 minutes later and it started! Sounds like the proxy might be spamming “start machine” requests a bit too often if there’s a lot of connections coming in?
We’ve been having some capacity issues and the problem is worse with volumes. A stopped machine does not presently reserve its resources. That means a start is not guaranteed to work on a machine if the server has already allocated all its memory.
We’re working on the start spam , however the real reason this worked later is because I rescheduled instances that don’t have volumes to other hosts (we have capacity in CDG, but the particular server you’re on was crowded).